fix: report partial and unresponsive profiling states

[marcopiraccini]

Summary

Profiling state reporting previously treated pod-side intent as actual state and could stall indefinitely when a worker event loop was blocked. This could make partial activation appear successful or cause the pod state to expire from ICC.

This change:

• reports the real isCapturing state for every worker and profile type
• returns a failed toggle result when one or more workers fail to start or stop profiling
• queries worker states in parallel with a configurable timeout
• coalesces concurrent reports so slow workers or scaler requests do not create overlapping runs
• applies bounded backoff to workers whose state queries time out
• keeps timed-out workers in reports as unresponsive, allowing ICC to show Pending, Partial, or Unknown instead of treating the healthy subset as the whole pod
• clears worker suppression when the worker restarts

Configuration

Adds PLT_FLAMEGRAPHS_STATE_QUERY_TIMEOUT, in milliseconds, with a default of 5000.